[FEAT]: Allow overriding compiler's chosen GPU arch via cmake

[jeremyfelder]

Description

There should be a flexible way to provide/override the compiler's chosen arch. Maybe the following would be best for that:

if(DEFINED CUDA_ARCH) // user defined arch takes priority
    set(CMAKE_CUDA_ARCHITECTURES ${CUDA_ARCH})
elseif(${CMAKE_VERSION} VERSION_GREATER_EQUAL "3.24.0") // otherwise, use native to detect GPU arch
    set(CMAKE_CUDA_ARCHITECTURES native)
endif() // otherwise fallback to compiler's chosen arch

See #159 and #163 for additional details

Motivation

We should have a flexible way to provide/override the compiler's chosen arch.